NVIDIA Shows Interactive Ray Tracing On GPUs 260
MojoKid writes "During SIGGRAPH 2008 in Los Angeles, NVIDIA is demonstrating a fully interactive GPU-based ray tracer. The demo is based purely on NVIDIA GPU technology, and according to NVIDIA the ray tracer shows linear scaling during rendering of a complex, two-million polygon, anti-aliased automotive styling application. The article reproduces screenshots from NVIDIA's demo. At three bounces (rays being traced as they bounce three times through a scene), performance is demonstrated at up to 30fps at HD resolutions of 1920x1080 for an image-based lighting paint shader, ray-traced shadows, reflections and refractions running on four next-generation Quadro GPUs in an NVIDIA Quadro Plex 2100 D4 Visual Computing System." Meanwhile reader arcticstoat passes on Intel's latest claim that rasterisation will die out the next few years, possibly in favour of ray tracing.
Re:That's a nice canned post ya got there (Score:5, Informative)
The kind with a star next to his name, obviously.
don't quit your day job quite yet (Score:5, Informative)
The devil is in the details. Ray tracing with glossy surfaces is relatively easy. But if you want to simulate real-world textures like orange-peel, bark, hair, or skin, things can really slow down.
Re:That's a nice canned post ya got there (Score:5, Informative)
Story submitted: 11:00am. Your post submitted: 11:00am. There's just no way in hell you formulated a response and typed out all of that in less than a minute. So just what kind of douchebag are you, anyway?
Uh, maybe he looked at the story on the firehose.
Re:Beautiful (Score:4, Informative)
Sure you can. A human instinctively knows when something looks "right" or "wrong". And one of the reasons why rasterization is capped is due to lighting problems. Lighting technology has improved significantly in the last decade, but still not sufficiently to compete with raytracing. Raytraced lighting will look more natural to an untrained viewer.
Rasterization is heavy on hardware features to improve the quality and performance of the scene render. e.g. Blending, pixel shaders, z-buffers, etc. Ray tracing is a far simpler operation on the hardware side, though it still behooves the software side to improve the number of objects tested for rendering. (Nothing new there.)
Re:Beautiful (Score:5, Informative)
Every bounce casts a new ray, so "3 casts per pixel" is an accurate description.
Re:What a waste of resources (Score:5, Informative)
The gameplay improvement is in deformable physical environments. Combined with mainstream physics engines, raytracing would allow for a sea-change in gameplay by allowing interactive gaming environments.
Raster methods rely on a bunch of tricks, many of which need to be precalculated for static maps. The most obvious example is binary space partioning tables. This leads to very static feeling environments that disallow interaction beyond doors of various types and moving platforms.
Re:Beautiful (Score:5, Informative)
The screenshots look relatively ugly because of the hashed-together-demo quality of the environment textures. But it's not a texture demo, it's a raytraced lighting demo.
Bare in mind this is ray tracing at a very rough and ready stage, but the potential is enormous. If you want to see the sort of effects it can achieve, check out some professional 3DSMax/VRay renders.
There's a nice render here [wikimedia.org] for illustrative purposes.
That's just a single frame with high quality textures, but it surely shows the potential.
Re:Beautiful (Score:5, Informative)
It isn't. More casts per pixel means that more rays are cast at slightly different angles through each pixel, and those are then averaged to yield the actual pixel colours.
Three bounces per ray simply means that a single ray can bounce three times before it's colour values are known.
Re:Not quite yet, I'm affraid (Score:2, Informative)
I.e., ray tracing looks grrreat and is the cheapest for shiny cars, crystal cups, and the like. Which is why everyone ray-traces cars and the like. It sucks for something like human skin, unless, of course, you want to make those humans look like polished shiny plastic dolls.
Subsurface Scattering.
http://en.wikipedia.org/wiki/Subsurface_scattering
http://graphics.ucsd.edu/~henrik/images/imgs/layered_skin_model.jpg
Re:Not quite yet, I'm affraid (Score:1, Informative)
I imagine that all the examples use highly reflective materials to dramatically show it's accuracy in the number of bounces being made -- not because that's the only thing it's good at rendering.
I'm excited about ray-tracing for specular reflections and subsurface scattering. Obviously you don't want to render everything with mirror-like reflections but virtually everything has some level -- however slight -- of reflection. Also good fresnel reflection [increasing reflection at increasingly-oblique angles] will help bring realism to renderings.
Re:Beautiful (Score:4, Informative)
I used to hear exactly the same things being said by the ray-tracing evangelists in the FilmFX industry 15 years ago. Rasterization is still the primary techinique used for any film you care to mention, and I'm almost 100% certain it will still be the primary technique 30 years from now.
Re:Beautiful (Score:3, Informative)
No, no, no. You have one shadow map per light source in most implementations (or a cube map for a point light. Depends on the algorithm used). You render the scene (depth only) from the light's viewpoint into this texture, thus finding the world space coordinate of the front most geometry (the shadow casters). Then, when you render from the camera viewpoint you look up each pixel's world space coordinate in this texture to determine if it's behind a shadow caster or not. If it is, do nothing, else calculate light contribution for the pixel.
Again, look at Crysis. Every leaf of grass casts a shadow.
Re:Beautiful (Score:3, Informative)
hmm - I was a bit underwhelmed, myself. First of all, graphics people have done GPU ray tracing on small static scenes has been around since about 2003 or 2004. The real limitation then was available memory. This is a standard ray tracer (no photon mapping), and highlighting an optimal model (a reflective one, not a diffuse one). They mention it scales linearly, which means they replaced the fixed function pipeline and used only shaders.
The scene shown is 2 million polygons (about the baseline of this generation's hardware) but with a single light source and hard shadows. Three reflections is adequate to get a decent looking scene (I've done good looking scenes with 2 reflections, but it depends on the scene).